From dd12fccc938fba6ae89772a122b8fe8988243d5d Mon Sep 17 00:00:00 2001 From: xinhui pan Date: Wed, 28 Nov 2018 21:14:56 +0800 Subject: [PATCH 1516/2940] drm/amdgpu: enable ras on sdma4 register IH, enable ras features on sdma. create sysfs debugfs file for sdma. Signed-off-by: xinhui pan Signed-off-by: Feifei Xu Signed-off-by: Eric Huang Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 4 + drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 184 ++++++++++++++++++++++- 2 files changed, 187 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h index 79e89e3bebc4..e988c3f09ecc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h @@ -30,6 +30,8 @@ enum amdgpu_sdma_irq { AMDGPU_SDMA_IRQ_TRAP0 = 0, AMDGPU_SDMA_IRQ_TRAP1, + AMDGPU_SDMA_IRQ_ECC0, + AMDGPU_SDMA_IRQ_ECC1, AMDGPU_SDMA_IRQ_LAST }; @@ -49,9 +51,11 @@ struct amdgpu_sdma { struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES]; struct amdgpu_irq_src trap_irq; struct amdgpu_irq_src illegal_inst_irq; + struct amdgpu_irq_src ecc_irq; int num_instances; uint32_t srbm_soft_reset; bool has_page_queue; + struct ras_common_if *ras_if; }; /* diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c index d6547a093e67..c0e568b68ef6 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c @@ -41,6 +41,8 @@ #include "ivsrcid/sdma0/irqsrcs_sdma0_4_0.h" #include "ivsrcid/sdma1/irqsrcs_sdma1_4_0.h" +#include "amdgpu_ras.h" + MODULE_FIRMWARE("amdgpu/vega10_sdma.bin"); MODULE_FIRMWARE("amdgpu/vega10_sdma1.bin"); MODULE_FIRMWARE("amdgpu/vega12_sdma.bin"); @@ -1495,6 +1497,83 @@ static int sdma_v4_0_early_init(void *handle) return 0; } +static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry); + +static int sdma_v4_0_late_init(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct ras_common_if **ras_if = &adev->sdma.ras_if; + struct ras_ih_if ih_info = { + .cb = sdma_v4_0_process_ras_data_cb, + }; + struct ras_fs_if fs_info = { + .sysfs_name = "sdma_err_count", + .debugfs_name = "sdma_err_inject", + }; + struct ras_common_if ras_block = { + .block = AMDGPU_RAS_BLOCK__SDMA, + .type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE, + .sub_block_index = 0, + .name = "sdma", + }; + int r; + + if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) { + amdgpu_ras_feature_enable(adev, &ras_block, 0); + return 0; + } + + *ras_if = kmalloc(sizeof(**ras_if), GFP_KERNEL); + if (!*ras_if) + return -ENOMEM; + + **ras_if = ras_block; + + r = amdgpu_ras_feature_enable(adev, *ras_if, 1); + if (r) + goto feature; + + ih_info.head = **ras_if; + fs_info.head = **ras_if; + + r = amdgpu_ras_interrupt_add_handler(adev, &ih_info); + if (r) + goto interrupt; + + r = amdgpu_ras_debugfs_create(adev, &fs_info); + if (r) + goto debugfs; + + r = amdgpu_ras_sysfs_create(adev, &fs_info); + if (r) + goto sysfs; + + r = amdgpu_irq_get(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC0); + if (r) + goto irq; + + r = amdgpu_irq_get(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC1); + if (r) { + amdgpu_irq_put(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC0); + goto irq; + } + + return 0; +irq: + amdgpu_ras_sysfs_remove(adev, *ras_if); +sysfs: + amdgpu_ras_debugfs_remove(adev, *ras_if); +debugfs: + amdgpu_ras_interrupt_remove_handler(adev, &ih_info); +interrupt: + amdgpu_ras_feature_enable(adev, *ras_if, 0); +feature: + kfree(*ras_if); + *ras_if = NULL; + return -EINVAL; +} + static int sdma_v4_0_sw_init(void *handle) { struct amdgpu_ring *ring; @@ -1513,6 +1592,18 @@ static int sdma_v4_0_sw_init(void *handle) if (r) return r; + /* SDMA SRAM ECC event */ + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_SDMA0, SDMA0_4_0__SRCID__SDMA_SRAM_ECC, + &adev->sdma.ecc_irq); + if (r) + return r; + + /* SDMA SRAM ECC event */ + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_SDMA1, SDMA1_4_0__SRCID__SDMA_SRAM_ECC, + &adev->sdma.ecc_irq); + if (r) + return r; + for (i = 0; i < adev->sdma.num_instances; i++) { ring = &adev->sdma.instance[i].ring; ring->ring_obj = NULL; @@ -1563,6 +1654,22 @@ static int sdma_v4_0_sw_fini(void *handle) struct amdgpu_device *adev = (struct amdgpu_device *)handle; int i; + if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA) && + adev->sdma.ras_if) { + struct ras_common_if *ras_if = adev->sdma.ras_if; + struct ras_ih_if ih_info = { + .head = *ras_if, + }; + + /*remove fs first*/ + amdgpu_ras_debugfs_remove(adev, ras_if); + amdgpu_ras_sysfs_remove(adev, ras_if); + /*remove the IH*/ + amdgpu_ras_interrupt_remove_handler(adev, &ih_info); + amdgpu_ras_feature_enable(adev, ras_if, 0); + kfree(ras_if); + } + for (i = 0; i < adev->sdma.num_instances; i++) { amdgpu_ring_fini(&adev->sdma.instance[i].ring); if (adev->sdma.has_page_queue) @@ -1600,6 +1707,9 @@ static int sdma_v4_0_hw_fini(void *handle) if (amdgpu_sriov_vf(adev)) return 0; + amdgpu_irq_put(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC0); + amdgpu_irq_put(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC1); + sdma_v4_0_ctx_switch_enable(adev, false); sdma_v4_0_enable(adev, false); @@ -1716,6 +1826,50 @@ static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev, return 0; } +static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + uint32_t instance, err_source; + + switch (entry->client_id) { + case SOC15_IH_CLIENTID_SDMA0: + instance = 0; + break; + case SOC15_IH_CLIENTID_SDMA1: + instance = 1; + break; + default: + return 0; + } + + switch (entry->src_id) { + case SDMA0_4_0__SRCID__SDMA_SRAM_ECC: + err_source = 0; + break; + case SDMA0_4_0__SRCID__SDMA_ECC: + err_source = 1; + break; + default: + return 0; + } + + amdgpu_ras_reset_gpu(adev, 0); + + return AMDGPU_RAS_UE; +} + +static int sdma_v4_0_process_ecc_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + struct ras_dispatch_if ih_data = { + .head = *adev->sdma.ras_if, + .entry = entry, + }; + amdgpu_ras_interrupt_dispatch(adev, &ih_data); + return 0; +} + static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) @@ -1743,6 +1897,25 @@ static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev, return 0; } +static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + unsigned type, + enum amdgpu_interrupt_state state) +{ + u32 sdma_edc_config; + + u32 reg_offset = (type == AMDGPU_SDMA_IRQ_ECC0) ? + sdma_v4_0_get_reg_offset(adev, 0, mmSDMA0_EDC_CONFIG) : + sdma_v4_0_get_reg_offset(adev, 1, mmSDMA0_EDC_CONFIG); + + sdma_edc_config = RREG32(reg_offset); + sdma_edc_config = REG_SET_FIELD(sdma_edc_config, SDMA0_EDC_CONFIG, ECC_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + WREG32(reg_offset, sdma_edc_config); + + return 0; +} + static void sdma_v4_0_update_medium_grain_clock_gating( struct amdgpu_device *adev, bool enable) @@ -1908,7 +2081,7 @@ static void sdma_v4_0_get_clockgating_state(void *handle, u32 *flags) const struct amd_ip_funcs sdma_v4_0_ip_funcs = { .name = "sdma_v4_0", .early_init = sdma_v4_0_early_init, - .late_init = NULL, + .late_init = sdma_v4_0_late_init, .sw_init = sdma_v4_0_sw_init, .sw_fini = sdma_v4_0_sw_fini, .hw_init = sdma_v4_0_hw_init, @@ -2010,11 +2183,20 @@ static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = { .process = sdma_v4_0_process_illegal_inst_irq, }; +static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = { + .set = sdma_v4_0_set_ecc_irq_state, + .process = sdma_v4_0_process_ecc_irq, +}; + + + static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev) { adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs; adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs; + adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_LAST; + adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs; } /** -- 2.17.1